Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
1. Claims 1-2, 5, 7-9, 12, 14-16, and 19 are rejected under 35 U.S.C. 103 as being unpatentable over Zhang et al., “A Knowledge Graph based Approach for Mobile Application Recommendation,” arXiv:2009.08621v1 [cs.IR] 18 Sep 2020 in view of Chen et al., “Mobile App Tagging,” WSDM’16, February 22–25, 2016, San Francisco, CA, USA and Lombriser et al. (US 2015/0112818 A1) (“Lombriser”).
As to claim 1, Zhang teaches a method for augmenting a graph based on a new application, [Abstract: “With the rapid prevalence of mobile devices and the dramatic proliferation of mobile applications (apps), app recommendation becomes an emergent task that would benefit both app users and stock-holders. …Specifically, we first designed a knowledge graph construction method to model the user and app side information.” In regards to the limitation of “new,” § 3.1 teaches that “ARKG construction mainly involves 2 sub-tasks, i.e., entity identification and relation extraction.” Therefore, the application is “new” in the sense that is a newly identified entity. Note that as shown in Table 1, an application (App) is a type of head entity.] the method comprising:
extracting signals [§ 2, paragraph 3: “Definition 1 (Content-Topic Entity). Considering Readme texts of apps provided by developers contain rich app profiles and are crucial to the efficacy of the ARKG to do recommendation, We used probabilistic topic modeling to incorporate them into the ARKG. A Content-Topic entity is a distribution over terms, which can be used to explore users preference on specific topics.” Page 5 (§ 3.1), item 1: “Taking the Readme text of each app as a document, we can get a corpus, i.e., a collection of |A| documents. We first preprocessed the corpus, i.e., conducting tokenization, stop words removing, stemming, lemmatization and typo corrections.” That is, the Readme text and their derived data, such as the corpus and the tokens, are extracted. Additionally, side information also constitutes extracted signals (see § 2, paragraph 2: “addition, we have rich side information for users and apps (e.g., app attributes and description texts).”]
inputting the extracted signals into a machine learning model; [Page 5 (§ 3.1), item 2: “LDA model learning. Given a preprocessed corpus of documents, we used variational EM algorithm to estimate parameters in LDA model. Then for each app a, let Z = z1; z2; ; zK be the set of latent topics, we can obtain the parameters θa = θa1; θa2; ; θaK of its Dirichlet distribution over Z.” Note that “LDA” (Latent Dirichlet Allocation) is a machine learning model.]
receiving, […], a set of tags that correspond to the new application […]; [§ 2.1, paragraph 4: “Based on these kinds of entities, 18 relations were defined for the ARKG, the detailed information of which is listed in Table 1.” As shown in Table 1, each application “app” can be linked to a tag such as a Content-Topic tag, or a Category tag.]
updating the graph to include one or more nodes corresponding to the new application, with each tag in the set of tags linked to the one or more nodes with an edge, […] [§ 1, paragraph 3: [“To meet the above challenge, we proposed a KG based app recommendation approach. A KG is a type of directed heterogeneous graph in which nodes correspond to entities and edges correspond to relations[11]. Among various types of side information, the KG contains much more fruitful facts and introduces semantic relatedness among apps, which can help find their latent connections. Beyond that, the KG consists of relations with various types, which is helpful for exploring a users interests reasonably.” Note that “ARKG Construction” is taught in § 3.1, which teaches “entity identification” and “relation extraction.” Note that in this case, “building” includes the act of updating during the process of building, in the absence of further limitations precisely defining a specific pre-update state.]
receiving a query corresponding to a tag in the set of tags; [§ 4.2, paragraph 1: “To evaluate the performance of the proposed model, we compared it with the following representative baselines.” That is, the testing of the recommendation model is described here, wherein the built model is used in a prediction setting. The prediction process is described in § 3.2, last paragraph 1 “This component embeds the entities and relations on the granularity of triples. After getting its outputs, we can use them to make app recommendation directly and roughly by equation (5).” That is, referring equation (5), the probability is given as an input query of h, r, and t, where h is the head entity (an app, as shown in Table 1), r is the relation, and t is the tail entity (tags). See § 3.4, paragraph 1: “So for user u and app a, we concatenated the representations of the two components into a single vector to do prediction as follows.” See also FIG. 1, which teaches the predicted probabilities.].
providing a response to the query based on the one or more nodes corresponding to the new application. [As noted above, the “prediction” or “app recommendation” corresponds to a response to the query. Furthermore, this response is based on the graph as shown in FIG. 1, since the embedding model uses the knowledge graph. See 3.3: “Next, focused on app recommendation, we built upon the architecture of graph convolution network to further capture both high-order structure and semantic information in the ARKG to make more precise recommendation. Here we started by describing a single layer, and then discussed how to generalize it to multiple layers. As illustrated in Fig. 2 (a), one single layer mainly involves 2 steps: 1) for each entity, aggregating information from its neighbors to form its neighbors' aggregated vector; 2) integrating with its own current latent vector to update its embedding for the next layer.”]
Zhang does not explicitly teach:
(1) the limitation that the signals are extracted “from the new application”;
(2) the set of tabs being received “as output from the machine learning model”; [The Examiner notes that Zhang teaches part of this limitation in that the Content-Topic tag is determined by a machine learning model. However, Zhang does not teach the entirety of the limitation that the set of tags (plural) is output from the machine learning model.]
(3) Receiving “levels of confidence for each tag in the set of tags” and the related limitation of “the edge having a weight corresponding to the level of confidence of the tag.”
Chen teaches extracting signals “from the new application” [In general, this reference teaches extracting apps for new applications. See § 4.4 (“App Tag Extraction Approach”), paragraph 1: “Given a novel app a0 (without tags) as a query, we first apply the learned app similarity function to find a set of N nearest neighbor apps for this app via searching a large app database.” This uses images from the app’s interface. See § 1, paragraph 1: “App market is a new form of software repository which contains a wealth of multi-modal data associated with apps, e.g., text descriptions, screenshot images, user reviews, and so on.” See examples in FIG. 2, which shows interface screenshots of the “YouTube” app. Features are extracted from images as disclosed in § 4.2.3: “We apply the bag-of-(visual)-words (BoW) to represent an app in visual space. We extract SIFT features from images, cluster the SIFT features using the fast K-means algorithm.”] and receiving a set of tags “as output from the machine learning model” [Paragraphs 1-2: “Given a novel app a0 (without tags) as a query, we first apply the learned app similarity function to find a set of N nearest neighbor apps for this app via searching a large app database… Our objective is to automatically discover a list of tags associated with a0 by mining the “Description" and “Update" text of all the apps in A0 (A0:text in short) …To achieve this goal, we propose an unsupervised App Tag Extraction (ATE) approach… In the last step, we refine those terms with the highest TF-IDF scores by using a modified TextRank [13] algorithm.” That is, the entire tag extraction process uses a machine learning model that includes a K-means algorithm and an unsupervised machine learning model that uses learned latent vector representations (see § 4.3.3, paragraph 3: “Given a large auxiliary corpus, we can learn the vector representations of words in an unsupervised manner by utilizing the skip-gram architecture”).] and “levels of confidence for each tag in the set of tags” [§ 4.4, paragraph 2: “Then, the second step computes a TF-IDF score for each of the remaining terms.” This measure is defined in § 4.4.2 (“TF-IDF Weighting”), and is a confidence weight in which the highest scores are regarded the most confident.]
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have combined the teachings of Zhang with the teachings of Chen by extracting features from the application, such as features in screenshot images, and implementing a machine learning model that outputs a set of tags and associated levels of confidence, so as to arrive at the above discussed features of the claimed invention. The motivation would have been to produce tags that utilize multi-modal data in app markets (see Chen, abstract: “To address this problem, we propose a novel auto mobile app tagging framework for annotating a given mobile app automatically, which is based on a search-based annotation paradigm powered by machine learning techniques.” § 1, paragraph 1: “App market is a new form of software repository which contains a wealth of multi-modal data associated with apps, e.g., text descriptions, screenshot images, user reviews, and so on.”).
The combination of references thus far does not teach the related limitation of “the edge having a weight corresponding to the level of confidence of the tag.”
Lombriser teaches “the edge having a weight corresponding to the level of confidence of the tag.” [[0030]: “The related entity selector 120 may present all available relations dimensions for an entity set, or, alternatively, may present a proper subset of relationship dimensions. The proper subset may be suggested based on dimensional criteria, such as strongest relationships as indicated by an edge weight, a maximum node traversal in an entity relation graph, etc.” [0050]: “The order of the dimensions can also be based on edge weights (if included in the knowledge graph 200) that indicate a confidence in the accuracy of the relationship. For example, a relationship dimension corresponding to an edge weight of 0.98 would be rated higher than a relationship dimension corresponding to an edge weight of 0.58.”]
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have combined the teachings of the references combined thus far with the teachings of Lombriser by implementing the graph such that the edges have a weight corresponding to the level of confidence of the tag, so as to arrive at the claimed invention. Doing so would have enabled the graph to indicate a confidence in the accuracy of a relationship, which can be used to select entities based on the graph, as suggested by Lombriser (see parts quoted above).
As to claim 2, the combination of Zhang, Chen, and Lombriser teaches the method of claim 1, as set forth above.
Chen further teaches “wherein extracting signals from the new application comprises extracting signals from a graphical user interface of the new application and from a page on an application store corresponding to the new application.” [See § 1, paragraph 1: “App market is a new form of software repository which contains a wealth of multi-modal data associated with apps, e.g., text descriptions, screenshot images, user reviews, and so on.” See examples in FIG. 2, which shows interface screenshots of the “YouTube” app (see caption: “Example of the multi-modal information associated with the “YouTube" app in Google Play. The names of modalities are in red colour.”). In this example, “Google Play” is an app store with the shown information, including name, description, user reviews, and screenshots.]
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have combined the teachings of the references combined thus far, including the above teachings of Chen, so as to have arrived at the claimed invention of the instant dependent claim. The motivation for doing so is covered by the motivation given for Chen in the rejection of the parent independent claim.
As to claim 5, the combination of Zhang, Chen, and Lombriser teaches the method of claim 1, as set forth above.
Chen teaches “further comprising selecting the machine learning model from a plurality of candidate machine learning models, each of the candidate machine learning models trained on signals corresponding to a different type of application.” [§ 4.2.2 teaches the use of LDA model to process textdata (i.e., signals that are text), while § 4.2.3 as discussed above teaches the use of bag-of-(visual)-words (BoW) and K-means (“We apply the bag-of-(visual)-words (BoW) to represent an app in visual space. We extract SIFT features from images, cluster the SIFT features using the fast K-means algorithm… and finally compute a compact BoW representation”). Here, BoW and the K-means classifier are both trained machine learning models and collectively constitute a machine learning model for the image modality, different from the LDA model for the text modality. In other words, depending on the modality of the data, a corresponding model (i.e., candidate machine learning model trained for different modalities) is selected for use. The selection of the model is implied by the disclosure of Chen since Chen teaches handling different modalities, and the instant claim does not require a specific methodology of selection.]
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have combined the teachings of the references combined thus far, including the above teachings of Chen, so as to have arrived at the claimed invention of the instant dependent claim. The motivation for doing so is covered by the motivation given for Chen in the rejection of the parent independent claim.
As to claim 7, the combination of Zhang, Chen, and Lombriser teaches the method of claim 1, wherein providing the response to the query comprises providing a subset of the graph comprising nodes connected to the tag by one or more edges. [Zhang, § 3.2, paragraph 2: “To be more specific, for each triplet (h; r; t) in the ARKG, it learns two vectors for the head entity h, tail entity t and relation r respectively.” That is, the embedding model learns embeddings for each triplet, which corresponds to a subset of the graph. The Examiner notes that the instant claim does not define what the subset is provided to. Therefore, providing the subset to learn a model that is used for providing the response to the query constitutes an act of providing to the extent required by the instant claim.]
As to claims 8-9, 12, and 14, these claims are directed to a computer-readable medium for performing operations that are the same or substantially the same as those of claims 1-2, 5, and 7, respectively. Therefore, the rejections made to claims 1-2, 5, and 7 are applied to claims 8-9, 12, and 14, respectively.
Furthermore, Zhang teaches “A non-transitory computer-readable medium comprising memory with instructions encoded thereon, the instructions, when executed by one or more processors, causing the one or more processors to perform operations, the instructions comprising instructions to” [Since the method of Zhang relies on complex algorithms and complex datasets (see § 4.1), it is understood that its method, as illustrated in FIG. 1, for example, is performed using a computer. Therefore, the instant limitations which are generic computer components are implied by the disclosure of Zhang.]
As to claims 15-16 and 19, these claims are directed to a computer-readable medium for performing operations that are the same or substantially the same as those of claims 1-2, 5, and 7, respectively. Therefore, the rejections made to claims 1-2, 5, and 7 are applied to claims 8-9, 12, and 14, respectively.
Furthermore, Zhang teaches “a system comprising: memory with instructions encoded thereon; and one or more processors that, when executing the instructions, are caused to perform operations comprising” [Since the method of Zhang relies on complex algorithms and complex datasets (see § 4.1), it is understood that its method, as illustrated in FIG. 1, for example, is performed using a computer. Therefore, the instant limitations which are generic computer components are implied by the disclosure of Zhang.]
2. Claims 3-4, 10-11, and 17-18 are rejected under 35 U.S.C. 103 as being unpatentable over Zhang in view of Chen and Lombriser, and further in view of Aggarwal et al. (US 2021/0326393 A1).
As to claim 3, the combination of Zhang, Chen, and Lombriser teaches the method of claim 1, as set forth above, but does not explicitly teach the further limitations of the instant claim.
Aggarwal teaches “wherein the machine learning model is trained with training data comprising a subset of the graph, the subset of the graph including one or more nodes labeled with tags. [[0035]: “In some cases, the graph 107 already exists and its construction need not be a concern. In addition to the nodes and edges, the graph 107 also includes embeddings (feature vectors) that encode information for each object and tag. The existing embeddings may also be ascertained a-priori and, together with the nodes and edges, comprise the training data for training model 110.” Note that the model in this case is used to determine tags, (see [0029]: “In one, the model may be used to predict tags for query objects….”) and is thus analogous to the model of the instant claim. [0034]: “model training component 101 trains model 110 jointly on graph 107, which includes nodes and edges corresponding to a dataset 105 of existing objects, tags, and their connections.”]
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have combined the references combine thus far with the teachings of Aggarwal by implementing the use of the graph for training data, so as to arrive at the claimed invention of the instant dependent claim. The motivation for doing so would have been to utilize existing objects and tags as information for training a model (see Aggarwal, [0034], part quoted above), since existing objects can be used to find similar objects to a query (see Aggarwal, [0007]: “In a brief example, a set of existing images may be queried to find images similar to a query image. In order to search the existing images, a graph is utilized to train a machine learning model on node embeddings.”).
As to claim 4, the combination of Zhang, Chen, and Lombriser teaches the method of claim 1, as set forth above, but does not explicitly teach the further limitations of the instant claim.
Aggarwal teaches “wherein the machine learning model is trained by: detecting a representative signal in the extracted signals; mapping the representative signal to a set of training data; and training the machine learning model with the set of training data.” [[0035]: “In some cases, the graph 107 already exists and its construction need not be a concern. In addition to the nodes and edges, the graph 107 also includes embeddings (feature vectors) that encode information for each object and tag. The existing embeddings may also be ascertained a-priori and, together with the nodes and edges, comprise the training data for training model 110.” Note that the model in this case is used to determine tags, (see [0029]: “In one, the model may be used to predict tags for query objects….”) and is thus analogous to the model of the instant claim. [0034]: “model training component 101 trains model 110 jointly on graph 107, which includes nodes and edges corresponding to a dataset 105 of existing objects, tags, and their connections.” In regards to the limitation of “detecting a representative signal in the extracted signals,” Zhang already teaches that the tags (tail entities in Table 1) are detected from the application data (extracted signals). Therefore, the signals that are formed into tags (which become nodes) are identified as representative, and such analogous tags are part of the training data in Aggarwal. Furthermore, the use of nodes that are tags for the training data constitutes mapping them to the training data.]
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have combined the references combine thus far with the teachings of Aggarwal by implementing the use of the graph for training data, so as to arrive at the claimed invention of the instant dependent claim. The motivation for doing so would have been to utilize existing objects and tags as information for training a model (see Aggarwal, [0034], part quoted above), since existing objects can be used to find similar objects to a query (see Aggarwal, [0007]: “In a brief example, a set of existing images may be queried to find images similar to a query image. In order to search the existing images, a graph is utilized to train a machine learning model on node embeddings.”).
As to claims 10-11, these claims recite further limitations that are the same or substantially the same as those of claims 3-4. Therefore, the rejections made to claims 3-4 are applied to claims 10-11.
As to claims 17-18, these claims recite further limitations that are the same or substantially the same as those of claims 3-4. Therefore, the rejections made to claims 3-4 are applied to claims 17-18.
3. Claims 6, 13, and 20 are rejected under 35 U.S.C. 103 as being unpatentable over Zhang in view of Chen and Lombriser, and further in view of Gurgul et al. (US 2023/0394770 A1) (“Gurgul”).
As to claim 6, the combination of Zhang, Chen, and Lombriser teaches the method of claim 1, as set forth above, but does not explicitly teach the further limitations of the instant dependent claim.
Gurgul teaches “further comprising filtering the set of tags by, for each tag in the set of tags: comparing the confidence level of the tag to a threshold confidence level; and responsive to the confidence level of the tag not exceeding the threshold confidence level, filtering the tag from the set of tags.” [[0131]: “The ranked tags 837 are an example of the objects 618, in accordance with some examples. The method 800 continues at operation 906 with filtering tags with confidence value less than a minimum threshold. The tags 624 are associated with a confidence value (not illustrated). The confidence value indicates a confidence that the tag 624 is relevant to the object 618 or that the tag 624 would be of interest to the user 646. The scan manager 924 removes tags 624 with an associated confidence value that is below a threshold value of confidence.”]
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have combined the teachings of the references combined thus far with the teachings of Gurgul by implementing the method to further comprise the filtering as recited in the instant claim, so as to arrive at the claimed invention of the instant denudement claims. The motivation would have been to remove tags that have low confidence, as suggested by Gurgul (see parts cited above).
As to claim 13, this claim recites further limitations that are the same or substantially the same as those of claim 6. Therefore, the rejection made to claims 6 is applied to claim 13.
As to claim 20, this claim recites further limitations that are the same or substantially the same as those of claim 6. Therefore, the rejection made to claims 6 is applied to claim 20.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure. The following documents depict the state of the art.
Mao et al., “Item Tagging for Information Retrieval: A Tripartite Graph Neural Network based Approach,” SIGIR ’20, July 25–30, 2020, Virtual Event, China teaches tagging content such as apps using a graph-based method.
Pujara et al., “Knowledge Graph Identification,” in H. Alani et al. (Eds.): ISWC 2013, Part I, LNCS 8218, pp. 542–557, 2013 teaches conventional techniques involving the building of knowledge graphs, including the use of confidence levels (known in this document as soft-truth values).
Alexander et al., US20210350252A1 teaches classifying software using knowledge graphs, including new applications (see [0030]-[0031]).
Motger et al., “Mobile Feature-Oriented Knowledge Base Generation Using Knowledge Graphs,” in A. Abelló et al. (Eds.): ADBIS 2023, CCIS 1850, pp. 269–279, 2023 teaches the use of tagging and knowledge graphs to catalog mobile apps.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to YAO DAVID HUANG whose telephone number is (571)270-1764. The examiner can normally be reached Monday - Friday 9:00 am - 5:30 pm.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Miranda Huang can be reached at (571) 270-7092. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/Y.D.H./Examiner, Art Unit 2124
/MIRANDA M HUANG/ Supervisory Patent Examiner, Art Unit 2124
Prosecution Insights